Fire on the Fringe: Empirically parameterizing defensible space behavior in an agent based model

Katie M. Lyon, Colorado State University
Michael Levy, University of California - Davis
Kenny Wallen, Texas A&M University
Patrick Bitterman, University of Iowa
Ellen Esch, University of California - San Diego
Michael Saha, University of Virginia
James D. Absher, USFS Pacific Southwest Research Station
Gerard Kyle, Texas A&M University

problem statement and/or research objective

• Wildfire and the WUI
• Social science & defensible space
• social behavior is contagious

• Humans are major driver of natural system dynamics, therefore models of natural systems need realistic representation of human behavior
• Traditional economic models impose strong assumptions about rationality and economically optimal behavior
• Evidence those assumptions are wrong and can bias model results

• ABM has advantage of flexible behavior.
• An agents behavior can be any function of
  • Environmental conditions
  • Social and economic conditions
  • Historical events
• However, that flexibility can be a liability: When behavior rules become even modestly complicated the drivers of system behavior can become elusive.
• Assumptions built into behavior rules
• Parameter space often too large to search completely
• So, need principled way to code behavior.
• Lots of thought on how to do it empirically (Smajgl book)
• Despite that, there is a well-developed, empirically grounded methodology for estimating behavior as a function of the kind of environmental variables that we find in a CHANS/SES model that is used in multiple disciplines: SEM.

1. Sample surveys
2. Participant observation
3. Field and laboratory experiments
4. Companion modeling
5. GIS and remotely sensed spatial data

Robinson et al., 2007

1. Sample surveys
2. Participant observation
3. Field and laboratory experiments
4. Companion modeling
5. GIS and remotely sensed spatial data

Robinson et al., 2007

Why SEM Examples of application to ABM

• Diffusion of Innovation (e.g, Lorscheid et al., 2014; Schwarz & Ernst, 2009)
• Theory of Planned Behavior (e.g., Ceschi et al., 2015)

• model household decisions of firewise behavior (i.e., do they create defensible space) in the San Diego WUI
• describe an empirical parameterization of agent behavior for a coupled model based on survey data of residents

• H1
• H2

• Empirical survey

Structural equation modeling

• standard statistical method in the social sciences

• Measurement model

  • relationships between latent variables and their observed indicators

• Structural model

  • relationships between the latent variables

SEM provides a modeling anchor for:
1. set of decision criteria that play a role in the given decision process
2. existing relationships between the criteria
3. relative relevance of criteria

Lorsheid et al., 2014

Cluster analysis Groups analysis

• Each agent gets a set of coefficient values from joint distribution
• Decision process model for agent
• For each agent typology?
• State variables change as a function of local environmental and social conditions in the model
• At each time step, agents update their likelihood of implementing DS behavior (implementation stochastic or deterministic?)

Decision criteria

• Simulate a fire
• Do they do defensible space? (yes/no)
  

• calibrate with survey data?

• Validation
• Scenarios

• Each household will make a decision about whether to implement (will assume that a change in behavior to firewise is permanent?)
• Will we assume a static population (i.e., without migration)?
• Fire season will occur at an annual timestep.

• What our model did (i.e., tell them what we told them we would do)
• What it did well? not well?
• Usefulness? Practicality?

Coupled model

• Figure of our coupled model?